outside of graphics, what other game stuff is ATi cards doing?
If you're going to test with gpu limited situations it shouldn't matter what the other game stuff is. Crazy tessellation like Aliens Vs Predator?
NVIDIA GF100 & Friends speculation
- Thread starter Arty
- Start date
outside of graphics, what other game stuff is ATi cards doing?
If you're going to test with gpu limited situations it shouldn't matter what the other game stuff is. Crazy tessellation like Aliens Vs Predator?
Well I think they just wanted to show tessellation performance. If nV's AA performance has gotten close to ATi's at x8 AA, bandwidth will become a predominant factor at those settings as does total memory amounts, so AA might cause an equalizing effect over all. Its all about the bottleneck right?
AA matters. Especially when it comes to running a game in highest resolution. It often becomes a bottleneck, especially when the cards memory isn't the highest. The performance isn't linear, so that you will see the impact of AA much more in some parts of the game then others.
Find an apple to apple benchmark were both cards runs 4X aa/16XAF @ 2048X1536 and you'll see the 8800 outperform the X1950XT in the whole game. AA matters.
There are different possible bottlenecks when doing graphics, you know?outside of graphics, what other game stuff is ATi cards doing?
What about taxing the texture units, the ROPs and the memory bandwidth a bit more than nvidia did in that heaven video (zero AA, no AF)? If you compare the heaven scores with tesselation off (or with a reduced amount of it), it appears the HD5870 likely ends up faster than the GTX470 or maybe even the GTX480.
No you missed the point why does the x1950xt loose over all (even with less AA) but in some instances is 3 times faster then the 8800 gts in the walk through. If it was AA that caused the problems you wouldn't see ATi's card shoot up like that, where the geforce's sit pretty even keel.
yes please read my post just two above yours, also there are parts of the scenes where tessellation is off and the geforce has a nice lead anyways not 2 times but closer to 20%. (the video does go along with the graph, when and where the tessellation is turned on and off.) The first 40 secs and the last 20 secs there really is no tessellation....
Where they turn on and off tessellation in the graph is where you see the first two big spikes also when you see the 45fps, that's where you see where the geforce has the 2 times lead (100 sec's or so)
Last edited by a moderator:
Look at the cyan curve of the 320mb 8800GTS on the same link approx 175sec to 191sec, or the beige 640mb 8800GTS from approx 203sec to 220sec:
http://www.hardocp.com/image.html?image=MTE3MDk3NTY2NDA3ZUxpb0ZKVm5fMTRfM19sLmdpZg==
Don't tell me that the different settings in AA and AF had nothing to do with this.
yeah a total of 30 seconds out of 260 secs
The point is that when there is no (or little) tesselation, both cards seem to be roughly on par, which clearly isn't good for NV.
Now, whether the GTX 480's advantage in tesselation makes a significant difference in actual games with "reasonable" amounts of tesselation (such as Stalker CoP or Alien vs Predator) remains to be seen, and will be a key issue.
The fact that NV keeps banging on the Heaven drum without mentioning pretty much anything else is rather worrying, though.
Well, as always, it would be a mistake to judge overall performance from one benchmark. We'll see how it looks when the card comes out. But my main point there was that if every benchmark for nVidia's new hardware basically looked like that (i.e. high minimum framerates), then it would be a fantastic card for everybody.
Now, of course, those minimum framerates may be entirely specific to tessellation, but we don't know that yet. Nor do we know if the higher framerates in the benchmark will bear any relation to how well the card does in other games.
why should it worry you, its not like they have ever given out this much information before a card was launched before have they? We would have been lucky to just get a picture of the card a couple of weeks before launch in the past. I think nV is using it because ATi was and had beat their chest with it for the HD5xxx series when it came to features and performance of Dx11 and their cards.
Look at the cards.
5850/5870 256 bit card with 1 gig of 1.2ghz(1ghz for the 5850) ram and a 2.1b transitor chip thats 330mm2 or there abouts ? and has been out since sept
The 470 on the other hand has a 320bit card with 1.5gigs of ram thats 1.2ghz downclocked to 800mhz as it seems that also requires more cooling and power .
If the 470 can be priced at $400 then ati can surely price the 5870 lower than its current cost.
IF the 470 falls between the 5850 and 5870 and doesn't fall inbetween those price points or worse is more expensive than the 5870 it could be alot of trouble for nvidia .
It will be interesting to see what happens. We really don't know the costs of each of these cards and what price ati can still make p rofit at .
You know what would be funny? If when Fermi finally comes out ATi raised the prices on their cards a bit...
about price if the gt300 is ~500mm, there is very little price advantage for ATi, its not like the rv770 and the gt200.
Memory differences between all three cards.
And do you think that 2X MSAA + 16AF or 4X SSAA + 16X AF or NOAA + 4X AF have no impact whatsoever in a scenario with different memory on a resolution of 2048X1536?
Edit:
This benchmark vs. the Heaven benchmark between 480 and the 5870 cannot be compared at all. I want the 480 to be as fast as it can be, but explaining away how close those cards were with little tessallation based on that old review is grasping for straws.
between the 320 and 640 mb versions of the 8800gts it makes a huge difference the x1950xt had 512 mb so no that's not where the problem is when you look at the x1950xt and 640 version of the 8800gts, since the x1950xt is still 3 times faster in some situations. So AA isn't causing any memory or page flipping since we would see the problem on the x1950xt first.
Long before anything about GF100 was known someone suggested here in the fora that Unigine's Heaven demo should be something like a industry standard. I still oppose to that notion; interpret and use a synthetic benchmarks only for their real purpose which sadly way too many abuse.
Wouldn't you think that a 320bit bus would rather suggest 10*128MB memory chips? And no the memory it contains isn't 1.2GHz.
The chip is still bigger. 170mm2 bigger if its just 500mm though some seem to think now thats its over 550mm.
Also you still have another half a gig on the cards that ati doesn't have to deal with and a 320bit bus.
*sigh*
care to go into more detail than just a sigh ?
Similar threads
- Deleted member 2197
- Architecture and Products
